Jingxiang Li

Research Interests

In recent years, the rapid development of embodied AI highlights the need for agents to autonomously perceive and interact in real-world environments, where high-precision, real-time 3D scene understanding is crucial. LiDAR, with its outstanding 3D spatial interpretability, has become a key sensing modality for this purpose. In large-scale, complex outdoor environments, efficient, accurate, and temporally consistent point cloud segmentation remains a core challenge for perception and mapping applications based on embodied AI.

My research addresses this challenge in three directions: developing robust segmentation models for outdoor LiDAR point clouds, integrating temporal information to improve sequential point cloud consistency, and applying knowledge distillation and model compression to build lightweight models. These approaches aim to enable embodied AI systems to achieve high-level perception in resource-constrained environments.

Education

• Ph.D. student in Geography and Remote Sensing, Department of Geography and Environmental Management, University of Waterloo, August 2025 – August 2029
• M.Sc. in Resources and Environment, School of Geomatics, Xi’an University of Science and Technology, September 2022 – June 2025
• Visiting M.Sc. Student, School of Computer Science, Universiti Sains Malaysia, March – September 2024

Publications

• Li Jingxiang, Tang F, Ma L, Zhu C, Gong Z, Ruhaiyem NIR, L J, 2025. Point-SCT: A multiscale spatial convolution-Swin Transformer network for point cloud ground filtering in complex mountainous terrains, IEEE Transactions on Geoscience and Remote Sensing, vol. 63, pp. 1-18, DOI: 10.1109/TGRS.2025.3573023.

• Li Jingxiang et al. “Multi-KPConv: Deep learning-based LiDAR Point Cloud Ground Point Extraction for complex terrains on the Loess Plateau, International Journal of Digital Earth, in press.